Introduction - clustering analysis of beer review dataset

This notebook performs an in-depth clustering analysis on a dataset containing various beer characteristics, including sensory ratings, chemical properties, and user reviews. The goal is to identify meaningful groupings of beers based on different subsets of features.

Key Steps Covered:

1. Optimal Parameter Selection

   • For KMeans, the Elbow Method and Silhouette Score are used to determine the best number of clusters.
   • For Gaussian Mixture Models (GMM), the Bayesian Information Criterion (BIC) is used.
   • For DBSCAN, the k-distance graph helps estimate an appropriate epsilon.

2. Model Comparison

   • The following clustering models are tested:
     KMeans, GaussianMixture, DBSCAN, HDBSCAN, MeanShift, AgglomerativeClustering.
   • Each model is evaluated using:
     • Silhouette Score (cluster separation)
     • Calinski-Harabasz Score (cluster compactness)
     • Davies-Bouldin Score (cluster similarity)

3. Visualization

   • PCA is used to project high-dimensional data into 2D space for visual inspection.
   • Radar plots visualize average feature values across clusters for:
     • Full feature set
     • Feature subsets: Sensory, Profile, Chemical, Reviews

4. Export

   • Radar plots are saved to projects/proj_3_team_5/plots/ for reporting and further analysis.

Importing and data load

import os
import sys
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from dotenv import load_dotenv
from sklearn.decomposition import PCA
from sklearn.neighbors import NearestNeighbors
from sklearn.cluster import KMeans, AgglomerativeClustering, DBSCAN, MeanShift, HDBSCAN
from sklearn.mixture import GaussianMixture
from sklearn.metrics import silhouette_score, calinski_harabasz_score, davies_bouldin_score
from kneed import KneeLocator

np.random.seed(42)

while any(marker in os.getcwd() for marker in ('exercises', 'notebooks', 'students', 'research', 'projects')):
    os.chdir("..") 
sys.path.append('.')

load_dotenv('projects/proj_3_team_5/.env')
df_path = os.getenv('PREPROCESSED_DATA_DIR')
df_cleaned_path = os.getenv('CLEANED_DATA_DIR')

df = pd.read_csv(df_path)
df_raw = pd.read_csv(df_cleaned_path)

Optimal parameter selection

# For KMeans, we use the Elbow method (inertia) and the Silhouette Score to select the optimal number of clusters.
# - The Elbow method helps identify the point where adding more clusters does not significantly reduce the within-cluster sum of squares (inertia), indicating a suitable number of clusters.
# - The Silhouette Score measures how similar an object is to its own cluster compared to other clusters, providing a quantitative metric for cluster quality.
# For GaussianMixture, we use the Bayesian Information Criterion (BIC) to select the optimal number of components.
# - BIC penalizes model complexity while rewarding goodness of fit, making it suitable for model selection in probabilistic clustering like GMM.
# - Lower BIC values indicate a better model, balancing fit and complexity.
inertia = []
silhouette = []
bic = []
n_range = range(2, 11)

for n in n_range:
    kmeans = KMeans(n_clusters=n, random_state=42)
    labels = kmeans.fit_predict(df)
    inertia.append(kmeans.inertia_)
    silhouette.append(silhouette_score(df, labels))
    
    gmm = GaussianMixture(n_components=n, covariance_type='full', random_state=42)
    gmm_labels = gmm.fit_predict(df)
    bic.append(gmm.bic(df))

# Plot KMeans elbow (inertia) and silhouette
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(n_range, inertia, marker='o')
plt.title('KMeans Elbow Method (Inertia)')
plt.xlabel('Number of clusters')
plt.ylabel('Inertia')

plt.subplot(1, 2, 2)
plt.plot(n_range, silhouette, marker='o')
plt.title('KMeans Silhouette Score')
plt.xlabel('Number of clusters')
plt.ylabel('Silhouette Score')
plt.tight_layout()
plt.show()

# Plot GMM BIC
plt.figure(figsize=(6, 4))
plt.plot(n_range, bic, marker='o')
plt.title('GaussianMixture BIC')
plt.xlabel('Number of components')
plt.ylabel('BIC')
plt.tight_layout()
plt.show()


# For DBSCAN, we use the k-distance graph to estimate the optimal value of eps (the neighborhood radius).
# - The k-distance plot helps visualize the distance to the k-th nearest neighbor for each point, sorted in ascending order.
# - The "elbow" in this plot suggests a threshold where points start to become outliers, which is a good candidate for eps.

k = 7
neigh = NearestNeighbors(n_neighbors=k)
nbrs = neigh.fit(df)
distances, indices = nbrs.kneighbors(df)
k_distances = np.sort(distances[:, k-1])

plt.figure(figsize=(6, 4))
plt.plot(k_distances)
plt.title('DBSCAN k-distance Graph')
plt.xlabel('Points sorted by distance')
plt.ylabel(f'{k}th Nearest Neighbor Distance')
plt.tight_layout()
plt.show()

Based on the KMeans parameter tuning:

• The Elbow Method shows a visible bend around $k = 8$, suggesting diminishing returns in inertia reduction beyond this point.
• However, the Silhouette Score is highest at $k = 2$ (approximately $0.12$), and drops sharply for higher values of $k$, even becoming negative, which indicates poor cluster separation.

Therefore, considering both methods, the optimal number of clusters is: $k=2$

optimal_kmeans = 2
optimal_gmm = 2
optimal_eps = 1250

For Agglomerative Clustering, we use the same number of clusters as determined optimal for KMeans, since both are hierarchical/partitioning methods and can be compared directly.

For MeanShift and HDBSCAN, we use their default parameter estimation, as these algorithms are designed to infer the number of clusters or density structure from the data.

PCA visualization and evaluation of clustering algorithms

models = {
    'KMeans': KMeans(n_clusters=optimal_kmeans, random_state=42),
    'GaussianMixture': GaussianMixture(n_components=optimal_gmm, covariance_type='full', random_state=42),
    'DBSCAN': DBSCAN(eps=optimal_eps, min_samples=k),
    'HDBSCAN': HDBSCAN(),
    'MeanShift': MeanShift(),
    'Agglomerative': AgglomerativeClustering(n_clusters=optimal_kmeans, linkage='complete')
}

pca = PCA(n_components=2)
X_pca = pca.fit_transform(df)

metrics = {}

plt.figure(figsize=(16, 12))
for i, (name, model) in enumerate(models.items(), 1):
    try:
        labels = model.fit_predict(df)
    except Exception as e:
        labels = np.zeros(df.shape[0])
        print(f"Model {name} failed: {e}")

    plt.subplot(3, 2, i)
    plt.scatter(X_pca[:, 0], X_pca[:, 1], c=labels, cmap='tab10', s=10)
    plt.title(f'{name} Clustering')
    plt.xlabel('PCA 1')
    plt.ylabel('PCA 2')
    n_clusters = len(set(labels)) - (1 if -1 in labels else 0)
    if n_clusters < 2:
        sil = -1
        ch = -1
        db = np.inf
    else:
        sil = silhouette_score(df, labels)
        ch = calinski_harabasz_score(df, labels)
        db = davies_bouldin_score(df, labels)
    metrics[name] = {
        'silhouette': sil,
        'calinski_harabasz': ch,
        'davies_bouldin': db,
        'n_clusters': n_clusters
    }

plt.tight_layout()
plt.show()

Selection of best clustering model

sensory_cols = ['review_aroma', 'review_appearance', 'review_palate', 'review_taste', 'review_overall']
profile_cols = ['Alcohol', 'Bitter', 'Sweet', 'Sour', 'Salty', 'Fruits', 'Hoppy', 'Spices', 'Malty', 'Astringency', 'Body']
chemical_cols = ['ABV', 'Min IBU', 'Max IBU']
review_cols = ['number_of_reviews']


metrics_df = pd.DataFrame(metrics).T
display(metrics_df)

# Choose the best clustering based on silhouette score (higher is better)
valid_metrics = metrics_df[metrics_df['n_clusters'] > 1]
if not valid_metrics.empty:
    best_model_name = valid_metrics['silhouette'].idxmax()
    print(f"Best clustering model: {best_model_name}")
else:
    best_model_name = metrics_df['silhouette'].idxmax()
    print(f"Best clustering model (by default): {best_model_name}")

best_model = models[best_model_name]
best_labels = best_model.fit_predict(df)

# Radar plot for the best clustering
radar_cols = sensory_cols + profile_cols + chemical_cols + review_cols
radar_cols = [col for col in radar_cols if col in df.columns]

# Compute mean for each cluster
cluster_means = pd.DataFrame(df[radar_cols])
cluster_means['cluster'] = best_labels
cluster_means = cluster_means.groupby('cluster').mean()

# Prepare radar plot
categories = radar_cols
N = len(categories)
angles = np.linspace(0, 2 * np.pi, N, endpoint=False).tolist()
angles += angles[:1]  # close the loop

plt.figure(figsize=(8, 8))
for idx, (cluster, row) in enumerate(cluster_means.iterrows()):
    values = row.values.flatten().tolist()
    values += values[:1]  # close the loop
    plt.polar(angles, values, label=f'Cluster {cluster}', linewidth=2)

plt.xticks(angles[:-1], categories, color='grey', size=10)
plt.title(f'Radar Plot of Cluster Means for {best_model_name}', size=15, y=1.08)
plt.legend(loc='upper right', bbox_to_anchor=(1.3, 1.1))
plt.tight_layout()
plt.savefig(f"projects/proj_3_team_5/plots/radar_{best_model_name}_overall.png", dpi=300)
plt.show()

# Add cluster labels to the original dataframe
df_with_clusters = df_raw.copy()
df_with_clusters['cluster'] = best_labels

# Show five sample beers from each cluster only if number of clusters is smaller than 10
n_clusters = len(df_with_clusters['cluster'].unique())
if n_clusters < 10:
    print(f"\n=== Sample Beers from Each Cluster ({best_model_name}) ===")
    for cluster in sorted(df_with_clusters['cluster'].unique()):
        cluster_beers = df_with_clusters[df_with_clusters['cluster'] == cluster]
        print(f"\nCluster {cluster} ({len(cluster_beers)} beers):")
        
        # Sample 5 beers from this cluster
        sample_beers = cluster_beers.sample(n=min(5, len(cluster_beers)), random_state=42)
        display(sample_beers)
else:
    print(f"\nSkipping sample display - too many clusters ({n_clusters})")

	silhouette	calinski_harabasz	davies_bouldin	n_clusters
KMeans	0.115463	193.196781	3.617928	2.0
GaussianMixture	0.115463	193.196781	3.617928	2.0
DBSCAN	-1.000000	-1.000000	inf	1.0
HDBSCAN	-0.093229	30.772725	3.962694	2.0
MeanShift	0.079885	2.358320	0.640568	410.0
Agglomerative	0.060042	87.414205	5.002694	2.0

Best clustering model: KMeans

=== Sample Beers from Each Cluster (KMeans) ===

Cluster 0 (1093 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
1219	Dogtoberfest	Lager - Märzen / Oktoberfest	Flying Dog Brewery	Flying Dog Brewery Dogtoberfest	There is sauerkraut in my lederhosen. I repeat...	5.6	18	25	17	46	...	69	10	115	3.479839	3.685484	3.527218	3.553427	3.639113	496	0
284	Mai-Ur-Bock	Bock - Maibock	Einbecker Brauhaus AG	Einbecker Brauhaus AG Einbecker Mai-Ur-Bock	“Ready for May?” In spring, the Einbecker brew...	6.5	20	38	14	33	...	74	10	112	3.688525	3.872951	3.827869	3.858607	3.862705	244	0
151	Pitchfork Rebellious Bitter	Bitter - English	RCH Brewery	RCH Brewery Pitchfork Rebellious Bitter	The name comes from the Pitchfork rebellion of...	4.3	20	35	34	49	...	112	12	64	3.423729	3.949153	3.627119	3.711864	3.889831	59	0
2026	Warlock	Stout - American Imperial	Southern Tier Brewing Company	Southern Tier Brewing Company Warlock	Imperial stout brewed with pumpkins Warlock is...	8.6	50	80	3	50	...	9	52	76	3.625000	4.125000	3.875000	4.000000	4.000000	4	0
1810	Gaelic Ale	Red Ale - American Amber / Red	Highland Brewing	Highland Brewing Highland Gaelic Ale	A deep amber colored American ale, featuring a...	5.8	25	45	7	23	...	74	4	57	3.665904	3.821510	3.778032	3.869565	3.964531	437	0

5 rows × 26 columns

Cluster 1 (1502 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
1893	Roggenbier	Rye Beer - Roggenbier	Real Ale Brewing Company	Real Ale Brewing Company Roggenbier	NaN	4.9	10	20	8	25	...	15	24	48	3.600000	3.900000	3.766667	3.733333	3.866667	15	1
1075	Löwenbräu Original	Lager - Helles	Löwenbräu AG	Löwenbräu AG Löwenbräu Original	NaN	5.2	18	25	22	29	...	60	7	56	3.097594	3.304813	3.391711	3.407754	3.616310	374	1
783	Deuchars IPA	IPA - English	The Caledonian Brewing Company	The Caledonian Brewing Company Deuchars IPA	4.4% ABV in bottles and 3.8% in cask.\t	4.4	35	60	27	53	...	90	18	59	3.782051	3.722222	3.735043	3.888889	4.021368	117	1
769	India Pale Ale	IPA - English	Meantime Brewing Company Limited	Meantime Brewing Company Limited India Pale Ale	NaN	7.5	35	60	16	42	...	99	11	84	3.847756	4.097756	3.956731	4.016026	4.028846	312	1
823	Sünner Kölsch	Kölsch	Gebr. Sünner GmbH & Co. KG	Gebr. Sünner GmbH & Co. KG Sünner Kölsch	NaN	4.8	18	25	37	32	...	71	4	69	3.607759	3.702586	3.745690	3.732759	4.051724	116	1

5 rows × 26 columns

Clustering analysis and visualization by feature groups

feature_groups = {
    'Sensory': sensory_cols,
    'Profile': profile_cols,
    'Chemical': chemical_cols,
    'Reviews': review_cols
}

for group_name, columns in feature_groups.items():
    metrics = {}
    print(f"\n=== Feature Group: {group_name} ===")
    print("Clustering results for each model:")
    X = df[columns].dropna()
    n_components = min(2, X.shape[0], X.shape[1])
    pca = PCA(n_components=n_components)
    X_pca = pca.fit_transform(X)

    plt.figure(figsize=(20, 12))
    plt.suptitle(f'Feature Group: {group_name}', fontsize=16)

    # Find optimal parameters for the models using the elbow method

    # KMeans: Find optimal number of clusters using elbow and silhouette score
    sse = []
    silhouette_scores = []
    k_range = range(2, min(11, X.shape[0]))
    for k in k_range:
        kmeans = KMeans(n_clusters=k, random_state=42)
        labels = kmeans.fit_predict(X)
        sse.append(kmeans.inertia_)
        # Only compute silhouette if more than 1 cluster
        if len(set(labels)) > 1:
            sil = silhouette_score(X, labels)
        else:
            sil = -1
        silhouette_scores.append(sil)
    # Find elbow point (simple heuristic: where the decrease sharply slows)
    if len(sse) > 2:
        elbow_k = k_range[np.argmin(np.diff(sse, 2)) + 1]
    else:
        elbow_k = k_range[0]
    # Find k with maximum silhouette score
    best_sil_k = k_range[np.argmax(silhouette_scores)]
    # Combine: pick k that is closest to elbow_k but also has high silhouette (within 90% of max)
    sil_threshold = 0.9 * max(silhouette_scores)
    candidate_ks = [k for k, sil in zip(k_range, silhouette_scores) if sil >= sil_threshold]
    if candidate_ks:
        optimal_kmeans = min(candidate_ks, key=lambda k: abs(k - elbow_k))
    else:
        optimal_kmeans = elbow_k

    # GaussianMixture: Use same optimal number of components as KMeans
    optimal_gmm = optimal_kmeans

    # DBSCAN: Find optimal eps using k-distance graph (elbow method)
    from sklearn.neighbors import NearestNeighbors
    neigh = NearestNeighbors(n_neighbors=2)
    nbrs = neigh.fit(X)
    distances, indices = nbrs.kneighbors(X)
    distances = np.sort(distances[:, 1])
    # Heuristic: take the point of maximum curvature as optimal eps

    kneedle = KneeLocator(range(len(distances)), distances, S=1.0, curve="convex", direction="increasing")
    optimal_eps = distances[kneedle.knee] if kneedle.knee is not None else np.percentile(distances, 90)
    k = 2  # min_samples

    # Agglomerative: Use same optimal number of clusters as KMeans
    optimal_agglom = optimal_kmeans

    models = {
        'KMeans': KMeans(n_clusters=optimal_kmeans, random_state=42),
        'GaussianMixture': GaussianMixture(n_components=optimal_gmm, covariance_type='full', random_state=42),
        'DBSCAN': DBSCAN(eps=optimal_eps, min_samples=k),
        'HDBSCAN': HDBSCAN(),
        'MeanShift': MeanShift(),
        'Agglomerative': AgglomerativeClustering(n_clusters=optimal_agglom, linkage='complete')
    }


    for i, (model_name, model) in enumerate(models.items(), 1):
        labels = model.fit_predict(X)

        plt.subplot(2, 3, i)
        if n_components == 1:
            plt.scatter(X_pca[:, 0], [0]*len(X_pca), c=labels, cmap='tab10', s=10)
            plt.xlabel('PCA 1')
            plt.ylabel('0 (no 2nd PCA component)')
        else:
            plt.scatter(X_pca[:, 0], X_pca[:, 1], c=labels, cmap='tab10', s=10)
            plt.xlabel('PCA 1')
            plt.ylabel('PCA 2')

        n_clusters = len(set(labels)) - (1 if -1 in labels else 0)
        if n_clusters < 2:
            sil = -1
            ch = -1
            db = np.inf
        else:
            sil = silhouette_score(X, labels)
            ch = calinski_harabasz_score(X, labels)
            db = davies_bouldin_score(X, labels)
    
        plt.title(f'{model_name}')
        metrics[model_name] = {
            'silhouette': sil,
            'calinski_harabasz': ch,
            'davies_bouldin': db,
            'n_clusters': n_clusters
        }

    plt.tight_layout()
    plt.show()

    print("Model metrics for this group:")
    metrics_df = pd.DataFrame(metrics).T
    display(metrics_df)

    # Plot radar plot for the best model in this group
    valid_metrics = metrics_df[metrics_df['n_clusters'] > 1]
    if not valid_metrics.empty:
        best_model_name = valid_metrics['silhouette'].idxmax()
    else:
        best_model_name = metrics_df['silhouette'].idxmax()
    print(f"Best clustering model for {group_name}: {best_model_name}")

    best_model = models[best_model_name]
    best_labels = best_model.fit_predict(X)

    # Compute mean for each cluster
    cluster_means = pd.DataFrame(X, columns=columns)
    cluster_means['cluster'] = best_labels
    cluster_means = cluster_means.groupby('cluster').mean()

    # Prepare radar plot
    categories = columns
    N = len(categories)
    angles = np.linspace(0, 2 * np.pi, N, endpoint=False).tolist()
    angles += angles[:1]  # close the loop

    plt.figure(figsize=(8, 8))
    for idx, (cluster, row) in enumerate(cluster_means.iterrows()):
        values = row.values.flatten().tolist()
        values += values[:1]  # close the loop
        plt.polar(angles, values, label=f'Cluster {cluster}', linewidth=2)

    plt.xticks(angles[:-1], categories, color='grey', size=10)
    plt.title(f'Radar Plot of Cluster Means for {best_model_name} ({group_name})', size=15, y=1.08)
    plt.legend(loc='upper right', bbox_to_anchor=(1.3, 1.1))
    plt.tight_layout()
    plt.savefig(f"projects/proj_3_team_5/plots/radar__{best_model_name}_{group_name.lower()}.png", dpi=300)
    plt.show()

    # Add cluster labels to the original dataframe
    df_with_clusters = df_raw.copy()
    df_with_clusters['cluster'] = best_labels

    # Show five sample beers from each cluster only if number of clusters is smaller than 10
    n_clusters = len(df_with_clusters['cluster'].unique())
    if n_clusters < 10:
        print(f"\n=== Sample Beers from Each Cluster ({best_model_name}) ===")
        for cluster in sorted(df_with_clusters['cluster'].unique()):
            cluster_beers = df_with_clusters[df_with_clusters['cluster'] == cluster]
            print(f"\nCluster {cluster} ({len(cluster_beers)} beers):")
            
            # Sample 5 beers from this cluster
            sample_beers = cluster_beers.sample(n=min(5, len(cluster_beers)), random_state=42)
            display(sample_beers)
    else:
        print(f"\nSkipping sample display - too many clusters ({n_clusters})")

=== Feature Group: Sensory ===
Clustering results for each model:

Model metrics for this group:

	silhouette	calinski_harabasz	davies_bouldin	n_clusters
KMeans	0.490976	3242.767989	0.769278	2.0
GaussianMixture	0.205715	71.971462	4.538508	2.0
DBSCAN	0.242044	3.420279	5.444213	3.0
HDBSCAN	-0.116752	60.261845	1.972067	4.0
MeanShift	0.237655	397.893592	1.191438	9.0
Agglomerative	0.512676	2762.645432	0.702818	2.0

Best clustering model for Sensory: Agglomerative

=== Sample Beers from Each Cluster (Agglomerative) ===

Cluster 0 (2104 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
1223	Munsterfest	Lager - Märzen / Oktoberfest	Three Floyds Brewing Co. & Brewpub	Three Floyds Brewing Co. & Brewpub Munsterfest	NaN	6.0	18	25	18	46	...	50	11	95	3.649733	3.679144	3.745989	3.748663	3.893048	187	0
395	George	Brown Ale - American	Hill Farmstead Brewery	Hill Farmstead Brewery George	George was our grandfather’s brother, and Hill...	6.0	25	45	14	68	...	67	7	133	4.000000	4.083333	4.000000	3.958333	3.916667	12	0
2084	Best Extra Stout	Stout - Foreign / Export	Coopers Brewery Limited	Coopers Brewery Limited Coopers Best Extra Stout	Now here's a beer with punch. Coopers Best Ext...	6.3	30	70	5	65	...	43	10	98	3.753482	3.905292	3.710306	3.892758	3.870474	359	0
327	Mountain Holidays In Vermont	Bock - Traditional	Rock Art Brewery	Rock Art Brewery Mountain Holidays In Vermont ...	NaN	5.8	20	30	12	74	...	47	26	119	3.701299	3.837662	3.805195	3.746753	3.824675	77	0
209	St. Feuillien Blonde	Blonde Ale - Belgian	Brasserie St. Feuillien	Brasserie St. Feuillien St. Feuillien Blonde	NaN	7.5	15	30	29	38	...	63	32	35	3.829787	3.904255	3.723404	3.904255	3.936170	47	0

5 rows × 26 columns

Cluster 1 (491 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
2279	Faithfull Ale	Strong Ale - Belgian Pale	Dogfish Head Brewery	Dogfish Head Brewery Faithfull Ale	Faithfull Ale is a celebration of Pearl Jam's ...	7.0	20	40	24	48	...	43	19	47	3.289286	3.632143	3.460714	3.278571	3.300000	140	1
683	Ta Henket	Herb and Spice Beer	Dogfish Head Brewery	Dogfish Head Brewery Ta Henket	For this ambitious liquid time capsule, we use...	4.5	5	40	7	25	...	55	26	60	3.250000	3.490385	3.471154	3.355769	3.365385	52	1
1900	Orkiszowe	Rye Beer - Roggenbier	Browar Kormoran	Browar Kormoran Orkiszowe	NaN	5.1	10	20	0	0	...	0	0	0	3.000000	3.000000	3.000000	3.000000	3.000000	1	1
2438	Benediktiner Weissbier Dunkel	Wheat Beer - Dunkelweizen	Klosterbrauerei Ettal / Ettaler Klosterbetrieb...	Klosterbrauerei Ettal / Ettaler Klosterbetrieb...	NaN	5.4	10	15	5	26	...	11	15	57	3.250000	3.750000	3.250000	3.500000	3.500000	2	1
1871	Station 33 Firehouse Red	Red Ale - Irish	North Country Brewing	North Country Brewing Station 33 Firehouse Red	NaN	5.5	20	30	18	47	...	44	2	114	3.147059	3.705882	3.411765	3.441176	3.411765	17	1

5 rows × 26 columns

=== Feature Group: Profile ===
Clustering results for each model:

Model metrics for this group:

	silhouette	calinski_harabasz	davies_bouldin	n_clusters
KMeans	0.207494	504.609572	1.465888	6.0
GaussianMixture	-0.012702	175.887219	2.307591	6.0
DBSCAN	-0.056300	13.582412	1.617519	5.0
HDBSCAN	-0.419559	11.526929	1.582308	21.0
MeanShift	-1.000000	-1.000000	inf	1.0
Agglomerative	0.132923	346.715836	1.797764	6.0

Best clustering model for Profile: KMeans

=== Sample Beers from Each Cluster (KMeans) ===

Cluster 0 (551 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
2525	Namaste White Belgian-Style Witbier	Wheat Beer - Witbier	Dogfish Head Brewery	Dogfish Head Brewery Namaste	A witbier bursting with good karma. Made with ...	4.8	10	20	14	29	...	32	29	32	3.990291	3.987055	3.896440	3.998382	4.085761	309	0
594	Malmgård Jouluolut	Farmhouse Ale - Sahti	Malmgårdin Panimo	Malmgårdin Panimo Malmgård Jouluolut	NaN	4.5	0	0	0	0	...	0	0	0	3.000000	3.500000	3.500000	2.500000	3.500000	1	0
811	Smetoniška Gira	Kvass	Vofas-Engelman	Vofas-Engelman Smetoniška Gira	NaN	1.2	0	0	2	4	...	1	1	16	3.166667	3.500000	3.666667	3.166667	3.500000	3	0
592	Finlandia Sahti	Farmhouse Ale - Sahti	Finlandia Sahti Ky	Finlandia Sahti Ky Finlandia Sahti	NaN	8.0	0	0	7	18	...	17	6	19	4.500000	3.000000	3.000000	3.500000	3.500000	2	0
1119	Organic Beer Shinshu Sansan	Lager - Japanese Rice	Yo-Ho Brewing Company	Yo-Ho Brewing Company Organic Beer Shinshu Sansan	NaN	5.0	6	18	7	7	...	15	1	18	3.333333	3.000000	3.000000	3.000000	3.500000	6	0

5 rows × 26 columns

Cluster 1 (294 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
559	Green's Endeavour Dubbel Dark Ale	Dubbel	Green's Gluten Free Beers	Green's Gluten Free Beers Green's Endeavour	NaN	7.0	15	30	13	42	...	27	16	64	2.935185	3.583333	2.481481	2.509259	2.509259	54	1
208	Blond	Blonde Ale - Belgian	Brasserie de l'Abbaye du Val-Dieu	Brasserie de l'Abbaye du Val-Dieu Val-Dieu Blond	NaN	6.0	15	30	33	35	...	61	33	56	3.771277	3.941489	3.840426	3.856383	3.984043	94	1
1355	Lou Pepe - Gueuze	Lambic - Gueuze	Brasserie Cantillon	Brasserie Cantillon Cantillon Lou Pepe - Gueuze	NaN	5.0	0	10	23	23	...	6	2	5	4.335938	4.113281	4.230469	4.378906	4.406250	128	1
2555	Consecration	Wild Ale	Russian River Brewing Company	Russian River Brewing Company Consecration	Dark ale aged in Cabernet Sauvignon barrels wi...	10.0	5	30	31	28	...	9	9	16	4.366114	4.091232	4.283768	4.473934	4.296801	844	1
2527	Blanche De Chambly	Wheat Beer - Witbier	Unibroue	Unibroue Blanche De Chambly	The Blanche de Chambly label features the icon...	5.0	10	20	17	33	...	42	34	52	3.805526	3.837407	3.756111	3.840064	3.971307	941	1

5 rows × 26 columns

Cluster 2 (576 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
1298	Rusty Chain	Lager - Vienna	Flying Bison Brewing Company	Flying Bison Brewing Company Rusty Chain	The #1 best selling local craft beer in Buffal...	5.2	15	30	14	45	...	48	7	80	3.250000	3.562500	3.625000	3.625000	3.625000	8	2
461	Pullman Nut Brown	Brown Ale - English	Flossmoor Station Restaurant & Brewery	Flossmoor Station Restaurant & Brewery Pullman...	A traditional english brown ale, very nutty ar...	6.0	15	25	8	96	...	28	4	177	3.983553	3.917763	3.960526	4.078947	4.046053	152	2
1819	Captain Sig's Northwestern Ale	Red Ale - American Amber / Red	Rogue Ales	Rogue Ales Captain Sig's Northwestern Ale	Label of 22oz bottle:10 Ingredients: Pale 2-ro...	6.2	25	45	6	55	...	107	9	112	3.759939	4.027523	3.808869	3.799694	3.844037	327	2
2145	Black Magic Stout	Stout - Irish Dry	Empire Brewing Company	Empire Brewing Company Black Magic Stout	A traditional dry Irish stout, carbonated with...	4.8	30	40	43	100	...	44	12	114	3.625000	4.375000	3.541667	3.625000	3.750000	12	2
2034	Obsidian Stout	Stout - American	Deschutes Brewery	Deschutes Brewery Obsidian Stout	Deep, robust and richly rewarding, this is bee...	6.4	35	60	9	72	...	67	6	132	4.077964	4.304768	4.121134	4.266753	4.250000	776	2

5 rows × 26 columns

Cluster 3 (564 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
1916	Rye IPA	Rye Beer	Black Market Brewing Co.	Black Market Brewing Co. Rye IPA	NaN	7.5	10	80	17	31	...	89	18	75	3.650000	3.750000	3.650000	3.700000	3.700000	10	3
1847	Lavery Imperial Red Ale	Red Ale - Imperial	Lavery Brewing Company	Lavery Brewing Company Lavery Imperial Red Ale	BIG. HOPPY. RED. Irish beer gone incognito! Ou...	8.2	55	85	16	37	...	76	2	44	3.666667	4.166667	3.833333	3.750000	3.833333	6	3
874	Molson Ice	Lager - Adjunct	Molson Coors Canada	Molson Coors Canada Molson Ice	NaN	5.6	8	18	21	18	...	38	3	55	2.309816	2.812883	2.684049	2.625767	2.815951	163	3
1857	O'Hara's Irish Red	Red Ale - Irish	Carlow Brewing Company	Carlow Brewing Company O'Hara's Irish Red	NaN	4.3	20	30	15	40	...	68	2	99	3.505952	3.830357	3.500000	3.511905	3.684524	168	3
1623	Evil Power	Pilsner - Imperial	Three Floyds Brewing Co. & Brewpub	Three Floyds Brewing Co. & Brewpub Evil Power	A fortified European-style Pilsner lagered to ...	7.2	30	65	37	50	...	100	9	82	3.428571	3.803571	3.607143	3.508929	3.464286	56	3

5 rows × 26 columns

Cluster 4 (209 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
544	Benediction	Dubbel	Russian River Brewing Company	Russian River Brewing Company Benediction	Brown in color, Benediction has notes aromas a...	6.75	15	30	11	47	...	32	27	58	3.960000	3.880000	4.120000	4.180000	4.280000	25	4
2455	Dancing Man	Wheat Beer - Hefeweizen	New Glarus Brewing Company	New Glarus Brewing Company Dancing Man Wheat	If you dream of wheat this brew will get your ...	7.20	10	15	13	46	...	16	69	59	4.249315	4.236986	4.228767	4.335616	4.371233	365	4
1770	UFO Pumpkin	Pumpkin Beer	Harpoon Brewery	Harpoon Brewery UFO Pumpkin	Imagine a pumpkin vine wound its way in a fiel...	5.90	5	70	9	37	...	27	75	59	3.703125	3.710938	3.562500	3.531250	3.570312	64	4
2566	Samuel Adams Old Fezziwig AleBoston Beer Compa...	Winter Warmer	Boston Beer Company (Samuel Adams)	Boston Beer Company (Samuel Adams) Samuel Adam...	Old Fezziwig, Rich & Sweet: Like the character...	5.90	35	50	6	35	...	10	64	78	3.777236	3.906911	3.747561	3.843089	3.810163	1230	4
1518	Hell's Belle	Pale Ale - Belgian	Big Boss Brewing	Big Boss Brewing Hell's Belle	NaN	7.00	20	30	19	36	...	44	40	46	3.522222	3.627778	3.605556	3.538889	3.650000	90	4

5 rows × 26 columns

Cluster 5 (401 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
1955	Railbender Ale	Scottish Ale	Erie Brewing Co.	Erie Brewing Co. Railbender Ale	Erie Brewing Company flagship beer features a ...	6.8	9	25	5	40	...	22	6	103	3.422559	3.594276	3.602694	3.624579	3.664983	297	5
1944	Heavy Horse Scotch Ale	Scotch Ale / Wee Heavy	Big Sky Brewing Company	Big Sky Brewing Company Heavy Horse Scotch Ale	NaN	6.7	25	35	12	61	...	37	14	119	3.675000	3.891667	3.741667	3.691667	3.716667	60	5
71	Holidale	Barleywine - American	Berkshire Brewing Company Inc.	Berkshire Brewing Company Inc. Holidale	NaN	9.5	60	100	15	72	...	69	27	107	3.831325	3.987952	3.906627	3.990964	3.891566	166	5
1467	Thanksgiving Ale	Old Ale	Mayflower Brewing Company	Mayflower Brewing Company Mayflower Thanksgivi...	The first and only perennial offering in our C...	6.7	30	65	18	69	...	51	53	151	3.826389	3.798611	3.881944	3.972222	3.972222	72	5
303	Southampton May Bock	Bock - Maibock	Southampton Publick House	Southampton Publick House Southampton May Bock	NaN	6.5	20	38	23	57	...	43	11	116	3.835227	3.880682	3.977273	4.113636	4.221591	88	5

5 rows × 26 columns

=== Feature Group: Chemical ===
Clustering results for each model:

Model metrics for this group:

	silhouette	calinski_harabasz	davies_bouldin	n_clusters
KMeans	0.515126	2293.056586	0.852369	2.0
GaussianMixture	0.469307	1347.788702	1.145447	2.0
DBSCAN	0.248944	312.481734	1.674888	15.0
HDBSCAN	0.429212	31.598181	1.986562	222.0
MeanShift	0.436874	1198.007788	0.772829	3.0
Agglomerative	0.442224	2029.390041	0.983633	2.0

Best clustering model for Chemical: KMeans

=== Sample Beers from Each Cluster (KMeans) ===

Cluster 0 (2072 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
1558	English Ale	Pale Ale - English	St. Peter's Brewery Co Ltd	St. Peter's Brewery Co Ltd St. Peter's English...	NaN	4.5	20	40	38	50	...	98	3	78	3.445455	3.543182	3.577273	3.615909	3.747727	220	0
906	Cypress Honey Lager	Lager - American Amber / Red	Granville Island Brewery	Granville Island Brewery Cypress Honey Lager	Brewed in small batches, our Cypress Honey Lag...	4.7	18	30	11	34	...	26	2	132	3.000000	3.250000	3.339286	3.125000	3.446429	28	0
851	Labatt Blue	Lager - Adjunct	Labatt Brewing Company Ltd.	Labatt Brewing Company Ltd. Labatt Blue	Labatt Blue is the best-selling Canadian beer ...	5.0	8	18	16	22	...	20	2	32	2.406096	2.690280	2.696870	2.672982	3.060956	607	0
1692	Bully! Porter	Porter - English	Boulevard Brewing Co.	Boulevard Brewing Co. Bully! Porter	The intense flavors of dark-roasted malt in Bo...	6.0	20	30	10	79	...	36	14	99	3.804393	4.125523	3.791841	3.991632	4.001046	478	0
482	Red Bird Ale	California Common / Steam Beer	Portsmouth Brewing Co. / Mault's Brewpub	Portsmouth Brewing Co. / Mault's Brewpub Ports...	An American Home Run! Named for the 1939 Ports...	4.8	35	45	8	22	...	26	3	30	3.333333	3.500000	3.250000	3.083333	3.250000	6	0

5 rows × 26 columns

Cluster 1 (523 beers):

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews	cluster
2593	Winter Shredder	Winter Warmer	Cisco Brewers Inc.	Cisco Brewers Inc. Winter Shredder	NaN	8.8	35	50	15	37	...	45	67	74	4.125000	3.875000	3.875000	3.750000	4.000000	4	1
2212	Life and Limb	Strong Ale - American	Sierra Nevada Brewing Co.	Sierra Nevada Brewing Co. Life & Limb	Brewed in collaboration with Dogfish Head Craf...	10.2	40	100	6	52	...	36	22	112	3.817507	4.114243	3.947329	3.996291	3.887240	674	1
42	Olde GnarlyWine	Barleywine - American	Lagunitas Brewing Company	Lagunitas Brewing Company Olde GnarlyWine	2011: 10.6% ABV, 69 IBU\t	10.9	60	100	14	49	...	68	17	89	4.052373	4.143208	4.082651	4.127660	4.016367	611	1
1450	Fourth Dementia - Bourbon Barrel-Aged	Old Ale	Kuhnhenn Brewing Company	Kuhnhenn Brewing Company Kuhnhenn Bourbon Barr...	This is our 4th Dementia Olde Ale that has bee...	13.5	30	65	13	61	...	18	32	124	4.555556	3.941919	4.340909	4.638889	4.474747	198	1
698	Furious	IPA - American	Surly Brewing Company	Surly Brewing Company Furious	A tempest on the tongue, or a moment of pure h...	6.7	50	70	11	24	...	96	3	61	4.374592	4.271650	4.200572	4.379902	4.336601	1224	1

5 rows × 26 columns

=== Feature Group: Reviews ===
Clustering results for each model:

Model metrics for this group:

	silhouette	calinski_harabasz	davies_bouldin	n_clusters
KMeans	0.683902	8787.141046	0.507592	3.0
GaussianMixture	0.503322	4169.878920	0.578554	3.0
DBSCAN	0.633867	354.255109	1.799954	20.0
HDBSCAN	0.754922	212.036369	1.638289	210.0
MeanShift	0.708887	3923.101971	0.412261	3.0
Agglomerative	0.709907	4861.545776	0.412262	3.0

Best clustering model for Reviews: HDBSCAN

/var/folders/w5/bqtvk3ss7gbf6xw9z7wqv6s80000gn/T/ipykernel_31732/1022216729.py:148: UserWarning: Tight layout not applied. The bottom and top margins cannot be made large enough to accommodate all Axes decorations.
  plt.tight_layout()

Skipping sample display - too many clusters (211)